Call the gpt-image-2 API with a Bearer token, get a hosted PNG back in about 45 seconds. Text-to-image and image-to-image share the same endpoint. 2 credits per image, no retainer, no SDK install. Create a key and curl your first render before your coffee lands.
Used by 10,000+ developers shipping with the gpt-image-2 API
One endpoint. Bearer auth. Hosted output. Nothing extra to wire up.
The gpt-image-2 API is a REST surface that turns a prompt (and up to four reference images) into a production-ready PNG. You POST to /api/v1/images/generate with a Bearer token, receive a task_id, and poll /api/v1/images/task/[id] until the image is ready. Every response is JSON — success envelope on 2xx, error envelope with a stable code and request_id on 4xx and 5xx. Each render is stored on our R2 CDN and served over HTTPS, so you can hotlink the URL or mirror it to your own bucket. Credits are deducted atomically at submit time and refunded automatically when a task fails, so your billing never drifts from your actual output. The surface is user-scoped: keys inherit the owning account's credit balance and can only see tasks created programmatically, not tasks from the web studio. Everything you send is JSON — no multipart uploads, no SDK lock-in, no hidden retries.
POST /api/v1/images/generate handles both text-to-image and image-to-image for the gpt-image-2 API. Reference images auto-switch the mode; the rest of the request body stays identical.
Authenticate with Authorization: Bearer YOUR_API_KEY. The gpt-image-2 API rejects keys in query strings, so your tokens never leak into CDN or proxy logs.
Every completed gpt-image-2 API task returns a stable https://file.gptimg.co URL. No signed-URL expiry, no download-and-rehost loop, no temporary links.
The gpt-image-2 API covers square, portrait, landscape, widescreen, and ultra-wide. Pick the ratio your platform needs and ship. No letterboxing step after generation.
If the gpt-image-2 API returns a failed task, the credit charge is refunded automatically. You never need to build a reconciliation script.
Every gpt-image-2 API response, success or error, carries an X-Request-Id header and a matching request_id in the body. Support tickets resolve in minutes instead of days.
One prompt endpoint, no infra to stand up, billing you can explain to finance
Ship a new image feature without provisioning GPU pods or negotiating a vendor contract. The gpt-image-2 API is a flat fee per render — 2 credits for a 1:1 image, linear on batch size. Drop a client into your backend with a single HTTP call, push the feature behind a flag, and let usage find your price point. Teams move from idea to staging in under an afternoon.
Image-to-image with up to four references, all over the same endpoint
Pass 1–4 https URLs in reference_images and the gpt-image-2 API switches into image-to-image mode automatically. Teams reach for it for style transfer, restyle, product swap, and identity-preserving edits. No separate edit endpoint, no multipart form, no base64 bloat — only http(s) URLs are accepted, so your request bodies stay small and your logs stay readable.
Stable error codes, request_id correlation, provider-hidden messages
Every gpt-image-2 API response carries a request_id you can quote in support tickets. Errors follow a uniform envelope with error.code, error.message, and error.request_id fields. A small set of stable codes covers every failure mode — UNAUTHORIZED, INVALID_REQUEST, INSUFFICIENT_CREDITS, NOT_FOUND, RATE_LIMITED, INTERNAL_ERROR. Upstream provider names are scrubbed from every message, so your clients never see leaked vendor internals when something breaks.
Flat-rate credits, R2 mirroring, and a single surface for both modes
The gpt-image-2 API gives you one flat credit price per render, hosted PNG URLs that do not expire in 24 hours, and a single endpoint that covers text-to-image and image-to-image. The same image through OpenAI's images endpoint costs $0.01–$0.20 depending on size and quality, with temporary signed URLs you have to download and rehost. Batch n=4 and pay linear credits — no separate endpoints, no hidden quality-tier pricing, no per-seat licensing.
Four steps. First curl in under five minutes.
Sign in, open Settings → API Keys, click Create, and copy the plaintext key once. The gpt-image-2 API shows the key only at creation time — store it in a secret manager.
Send JSON with prompt, aspect_ratio, n, and optional reference_images. The gpt-image-2 API returns a task_id and a processing status immediately.
Wait ten seconds after submission, then poll every five seconds until the gpt-image-2 API flips status to completed or failed. Typical end-to-end latency is 45–60 seconds.
Completed gpt-image-2 API responses contain a stable https://file.gptimg.co URL. Hotlink it, mirror it to your storage, or hand it straight to your frontend.
Integration, pricing, and rate limits
The gpt-image-2 API ships with free signup credits, a Bearer token in two clicks, and a 45-second round trip from prompt to hosted PNG. No SDK install, no credit card to start.